Build a Rails product

A senior Rails team designs and ships a new product or a v1 on Rails 8.1 with Ruby 3.4, or 4.0 where the app's own suite and latency measurements support it, using the framework's own defaults where they fit and replacing them only when we can say why. If the product needs a model, we build the Rails side and the inference service together, with the boundary between them written down before the first commit.

Sources: Rails 8.0 release notes; Rails 8.1 release post.

Founders and product leads who have validated the problem and need a v1 that a future in-house team can own. Teams adding a product line to an existing Rails business. Products where an LLM feature or a classical model is part of the value, not a demo.

Not for you if: Anyone still choosing between ideas; we can help with that conversation, but it is not a build. Anyone who wants a fixed price before the codebase exists. Anyone who needs a stack other than Rails; we will tell you so and stop.

A working session on the assumptions that would kill the product, then a short review of what exists (a prototype, a spreadsheet, an API contract). We agree the first release boundary, the data model, the test strategy for the money, authentication and personal-data paths, and the deploy target. Only then do we scaffold.

Rails 8.0 shipped a stack that needs no Redis and no PaaS to run: Solid Queue, Solid Cache, Solid Cable, Propshaft, Kamal 2 and Thruster, with 37signals reporting Solid Queue at 20 million jobs a day for HEY. Rails 8.1 added Active Job Continuations, structured event reporting and bin/ci with config/ci.rb. The Ruby SDKs for the major model providers are vendor maintained and current: the anthropic gem reached 1.68.0 on 1 September 2026 and the openai gem 0.85.0 on 3 September 2026, still pre-1.0.

The honest limit: Rails is used by 5.9 percent of respondents in the 2025 Stack Overflow survey against Node.js at 48.7 percent and Next.js at 20.8 percent. You are choosing a small, senior pool. That is the argument for a firm that has been in it since 2005. We do not publish a claim that Rails builds a first release faster than Next.js or Django; no measured comparison exists in any source we have opened, and we will not invent one.

Sources: Rails 8.0 release post; Rails 8.1 release post; anthropic gem; openai gem; Stack Overflow 2025.

Rails owns the jobs, retries, events, tenancy and UI. Python owns the training and the heavy inference. The GVL runs one thread of Ruby at a time per process, so model work sits in a service beside the monolith, which is how our own products are built: iPAM v2 runs Rails 8.1.3.1 services beside a Python ML service with MLflow, and NeuroSight runs a Rails alert manager beside YOLOv8 edge inference.

Sources: The Ruby GVL, Speedshop; Ractors on Rails, 11 August 2026.

Our own products marketed on this site are Rails builds with model workloads beside them. The public recommendations of our founder from 2007 to 2012, nineteen of them from clients, are almost all about builds delivered fast, clean and on the deadline. Case studies are anonymous on the site, with detail under NDA.